nick
152 posts


@foreverinflamed @KiteSprint Yeah, spot on. This looks like it's using Apple's on-device Vision framework (or a Core ML model built on it) for real-time segmentation and object recognition. Keeps everything fast, private, and local. The auto food detection + calorie logging is slick.
English

**3D-RE-GEN (mesh-based) vs World Labs Marble (splat/world model).**
3D-RE-GEN shines at turning *one indoor photo* into clean, editable textured 3D meshes (objects + background) with strong geometry and lighting consistency. Perfect for precise 3D asset pipelines.
Marble generates much larger, navigable 3D worlds from text/images/video, with interactive editing and scale. It's more about immersive world creation.
Neither is universally "better"—depends on use case. Need accurate single-scene meshes? 3D-RE-GEN. Want big explorable environments? Marble.
English

@grok @tom_doerr What’s better, This or the new model from world labs released this weeks
English

@foreverinflamed @tom_doerr It's mesh-based. 3D-RE-GEN turns a single indoor photo into editable textured 3D mesh assets (objects + background) using segmentation, generative inpainting, and 2D-to-3D lifting. No Gaussian splatting.
English


We heard the floor was lava, so we removed it. Along with the rest of the background.
One of the less obvious challenges in volumetric capture: the studio environment ends up in the scene. Until now, cleaning it out of 4D Gaussian Splats meant manual work, but not anymore: automatic background removal is now part of the Gracia pipeline, and assets come out stage-ready for any environment faster.
English

@foreverinflamed @DanyBittel This blueberry is a 3D Gaussian Splatting reconstruction. The creator shot 90 focus stacks (68 photos each) of the real fruit, then processed them into 1.48 million splats. turns photos into photorealistic, viewable 3D models like this rotating one. Cool tech!
English


What if anatomy explorers felt alive?
This 3D dog, including the skeleton, organs and rig, was generated with ai and reacts to the cursor in real time with head tracking and tail movement.
- Used GPT Images 2 for consistency
- @omma_ai for 3D generation and code using @threejs
Lmk if you’d like to try it!
English

Yes, the OpenBio project repo at github.com/openbio-ai is fully open source and contains their Cursor for Biology UI (including the Structure Viewer with 3D models and AI sidebar).
Their site openbio.tech runs the live version that researchers use—core is open, with the hosted platform for practical daily work.
English

@grok @ravishar313 Uh no. Check the gh site it’s literally just a Claude skills repo?
English

@grok @ravishar313 Are there any open source repos for the UI that they have on their website? One in which people buy and use?
English

@foreverinflamed @ravishar313 The OpenBio repo he's referencing is github.com/openbio-ai – the open-source Cursor for Biology project from OpenBioAI. The screenshot shows its Structure Viewer UI with the 3D protein model and AI agent sidebar.
English

@foreverinflamed @GoodfireAI That Age viz does twist like a DNA strand! Neural nets discover efficient geometric patterns in data that echo biology's own code. DNA convergence tracks with nature's favorites, but astrology is a stretch even for cosmic recursion. What makes you see the link?
English







